92243 - Weird Floating point calculation problem in JavaScript

Status: VERIFIED WORKSFORME

(Core :: JavaScript Engine, defect)

Description

[Håkon Innerdal]

Operating environment:
Mozilla version 0.9.2 (Tested using galeon front-end and embedded mozilla
engine) both compiled from CVS mozilla using tag MOZILLA_0_9_2_RELEASE, 
using glibc 2.2.3, gcc-3.0 (clean release versions, not pre-release nor patched)
on a linux 2.4.4 i686 box.

TestCase:
--
<FORM NAME="testForm">
<INPUT TYPE="text" NAME="test" VALUE="0" SIZE=20>
</FORM>
<SCRIPT>
document.testForm.test.value = 200.1 - 80;
</SCRIPT>
--
Expected result:
the value in input text-box should read 120.1

Actual results:
The value in the input -field is "<0.1" not "120.1" after running the code. At
one time, (restarted galeon from scratch) I actually got the correct value, but
reloading the page gave me the incorrect value, so this error may not always be
consistent. entering a calculation of 200.1 - 70 works, giving 130.1 as result. 
200.1 - 90 gives ";0.1"
200 - 90 gives correct answer 130.

also
20.1 + 90 gives ";0.1" and not 120.1

The testcode runs fine on netscape communicator 4.7 for linux.

Regards, Håkon Innerdal

Comment 1

[Håkon Innerdal]

This problem doesn't occur on RedHat 7.0 and pre built mozilla 0.9.2. 2001062823
The problem _may_ be related to either gcc-3.0 or glibc-2.2.3... or galeon.

Comment 2

[Phil Schwartau]

Assigning to khanson, cc'ing dbaron to see if either can reproduce this.
I have not been able to so far -

Comment 3

[Phil Schwartau]

Attachment: Håkon's HTML testcases

Comment 4

[Alex Vincent [:WeirdAl]]

wfm Mozilla/5.0 (Windows; U; Win98; en-US; rv:0.9.6+) Gecko/20011123

Comment 5

[Kenton Hanson (gone)]

I don't understand Alex's comment.

Comment 6

[Phil Schwartau]

"wfm" === "Worksforme"

Comment 7

[Phil Schwartau]

Here are my versions of Linux and gcc, that WORKFORME:

[] uname -a  
Linux attica 2.2.12-20smp #1 SMP Mon Sep 27 10:34:45 EDT 1999 i686 unknown

[] gcc -v
Reading specs from /usr/local/lib/gcc-lib/i686-pc-linux-gnu/2.95.3/specs
gcc version 2.95.3 20010315 (release)

Comment 8

[Kenton Hanson (gone)]

Targeting for the future.  I need to reproduce the error.

Comment 9

[Philip Van Bogaert]

I have a simular bug 

running the following code: alert(217728000000);

outputs "217727:00000"

on mozilla 1.2 win2000

Comment 10

[Phil Schwartau]

Philip: thanks for the info. The bug you are finding with 
Mozilla 1.2 Win2000 has recently been fixed: see bug 140852,
"String(819187200000) == '8191871:0000' in xpcshell, browser".

Håkon: that makes me wonder - with a current trunk build, are you
still getting the bug you reported above? Or is that now fixed, too?

Comment 11

[Philip Van Bogaert]

there is no xpcshell command in my version (offical release)
but alert(819187200000); in the javascript console gives me a value with ':'

var x = 819187200000;
alert(x); --> error
alert(x.toFixed()); --> ok
// note all the other number to string functions/methods that i tried 
// have this bug 

Comment 12

[Håkon Innerdal]

I have been unable to reproduce the bug on my current setup:

Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.3a) Gecko/20030207 <developer build>
compiled with gcc3.2

So I assume the bug has been fixed, at least for my testcase reported initially.

Comment 13

[Phil Schwartau]

Based on Håkon's findings, I'm going to mark this bug as WORKSFORME.

Philip: you don't get that error with an up-to-date build, do you?
I would expect you to see it with your version 1.2 build:

                    var x = 819187200000;
                    alert(x); --> error

Bug 140852 was only fixed on the trunk on 2003-02-17, so only builds
after that date will work correctly. Can you try that? The Mozilla
version number in Help > About Mozilla should be 1.4. Thanks -

Comment 14

[Phil Schwartau]

Marking Verified.

Philip - please reopen this bug if you still see the error with
an up-to-date build; thanks -

Comment 15

[Philip Van Bogaert]

indeed the bug appears to be solved in the latest build
;)